You are viewing the course site for a past offering of this course. The current offering may be found here.
Lecture 13: Global Illumination and Path Tracing (62)

It may be hard to compare between changing slides, but the diminishing impact of higher bounce computation makes me wonder when one should stop computing. Especially since this decision seems like a fairly subjective one, to consider whether or not more bounce adds more realism.


@tristanburke I think in graphics the "fast and looks good enough" solution is often preferable to the optimal solution (if optimality is even well-defined for the problem - you're right that optimality might even be subjective). For instance, in CG fluid simulations the most commonly used algorithm is a lossy approximation of real fluids, but looks good enough to humans.


interesting to note that increasing the number of bounces in a render is able to maintain image fidelity during brightening in a more believable way, while most image post processing brightening increases the brightness of every pixel equally, lending itself to a blown out/out of contrast image.


I'm curious to know how many bounces are used for modern rendering for realistic images. After ~3 I don't think I would really notice


I'm also curious about whether or not if there's a "standard" for how many bounces are usually applied for global illumination in most modern renderings for realistic images. In the examples given, the differences are quite clearly evident in the first few images with less bounces, since it's the first few applications of light bounced from other objects in the scene, but the effect seems to plateau as we reach higher numbers of bounces. Since more bounces would mean that the renderings are more computationally intensive anyways, is there a standard for this, or would this be something that's more "depends on the situation?"


Really cool recent demo from cryengine with everything running in real time!

Note how much indirect lighting is used on almost everything using reflections from water, materials, glass, etc.


According to a few articles I've seen online, the most common practice for looking at models is to simply add more direct sources of light so that the image comes out cleaner, as adding too many bounces to global illumination can create extra noise and adds way more processing time in comparison. The "whatever is enough" approach mentioned by Andrew is definitely what I would still designate as "standard" since every developer will have a different setup, but the cost of increasing bounces to beyond a degree like this can end up scaling dramatically and seems usually not worth pursuing.

You must be enrolled in the course to comment